Skip to content
Snippets Groups Projects

Compare revisions

Changes are shown as if the source revision was being merged into the target revision. Learn more about comparing revisions.

Source

Select target project
No results found

Target

Select target project
No results found
Show changes
Commits on Source (1530)
[flake8]
max-line-length=120
exclude=src/pystencils/jupyter.py,
src/pystencils/plot.py
src/pystencils/session.py
ignore = W293 W503 W291 C901 E741
src/pystencils/_version.py export-subst
__pycache__
.ipynb_checkpoints
.coverage*
*.pyc
*.vti
/build
/dist
*.egg-info
.cache
_build
/html_doc
/.idea
.vscode
.cache
_local_tmp
RELEASE-VERSION
test-report
src/pystencils/boundaries/createindexlistcython.c
src/pystencils/boundaries/createindexlistcython.*.so
tests/tmp
tests/var
tests/kerncraft_inputs/.2d-5pt.c_kerncraft/
tests/kerncraft_inputs/.3d-7pt.c_kerncraft/
report.xml
coverage_report/
# macOS
**/.DS_Store
*.uuid
stages:
- pretest
- test
- nightly
- docs
- deploy
# -------------------------- Templates ------------------------------------------------------------------------------------
# Base configuration for jobs meant to run at every commit
.every-commit:
rules:
- if: $CI_PIPELINE_SOURCE != "schedule"
# Configuration for jobs meant to run on each commit to pycodegen/pystencils/master
.every-commit-master:
rules:
- if: '$CI_PIPELINE_SOURCE != "schedule" && $CI_PROJECT_PATH == "pycodegen/pystencils" && $CI_COMMIT_BRANCH == "master"'
# Base configuration for jobs meant to run at a schedule
.scheduled:
rules:
- if: $CI_PIPELINE_SOURCE == "schedule"
# -------------------------- Tests ------------------------------------------------------------------------------------
# Normal test - runs on every commit all but "long run" tests
tests-and-coverage:
stage: pretest
extends: .every-commit
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/full:cupy12.3
before_script:
- pip install -e .
script:
- env
- pip list
- export NUM_CORES=$(nproc --all)
- mkdir -p ~/.config/matplotlib
- echo "backend:template" > ~/.config/matplotlib/matplotlibrc
- mkdir public
- pytest -v -n $NUM_CORES --cov-report html --cov-report xml --cov-report term --cov=. -m "not longrun" --html test-report/index.html --junitxml=report.xml
- python -m coverage xml
tags:
- docker
- cuda11
- AVX
coverage: /Total coverage:\s\d+.\d+\%/
artifacts:
when: always
paths:
- coverage_report
- test-report
reports:
coverage_report:
coverage_format: cobertura
path: coverage.xml
junit: report.xml
# Normal test with longruns
tests-and-coverage-with-longrun:
stage: test
when: manual
allow_failure: true
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/full:cupy12.3
before_script:
- pip install sympy --upgrade
- pip install -e .
script:
- env
- pip list
- export NUM_CORES=$(nproc --all)
- mkdir -p ~/.config/matplotlib
- echo "backend:template" > ~/.config/matplotlib/matplotlibrc
- mkdir public
- py.test -v -n $NUM_CORES
tags:
- docker
- cuda11
- AVX
# pipeline with latest python version
latest-python:
stage: test
extends: .every-commit
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/latest_python
before_script:
- pip install -e .
script:
- env
- pip list
- pip install -e .
- export NUM_CORES=$(nproc --all)
- mkdir -p ~/.config/matplotlib
- echo "backend:template" > ~/.config/matplotlib/matplotlibrc
- mkdir public
- py.test -v -n $NUM_CORES -m "not longrun" --junitxml=report.xml
tags:
- docker
- AVX
artifacts:
when: always
reports:
junit: report.xml
# Minimal tests in windows environment
#minimal-windows:
# stage: test
# tags:
# - win
# script:
# - export NUM_CORES=$(nproc --all)
# - source /cygdrive/c/Users/build/Miniconda3/Scripts/activate
# - source activate pystencils
# - pip install joblib
# - pip list
# - python -c "import numpy"
# - py.test -v -m "not (notebook or longrun)"
ubuntu:
stage: test
extends: .every-commit
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/ubuntu
before_script:
- ln -s /usr/include/locale.h /usr/include/xlocale.h
- pip3 install -e .
script:
- export NUM_CORES=$(nproc --all)
- mkdir -p ~/.config/matplotlib
- echo "backend:template" > ~/.config/matplotlib/matplotlibrc
- sed -i 's/--doctest-modules //g' pytest.ini
- env
- pip list
- pytest -v -n $NUM_CORES -m "not longrun" --junitxml=report.xml
tags:
- docker
- cuda11
- AVX
artifacts:
when: always
reports:
junit: report.xml
.multiarch_template:
stage: test
extends: .every-commit
allow_failure: true
before_script: &multiarch_before_script
# - pip3 install -v .
- export PYTHONPATH=src
- python3 -c "import pystencils as ps; ps.cpu.cpujit.read_config()"
- sed -i '/^fail_under.*/d' pytest.ini
script:
- export NUM_CORES=$(nproc --all)
- mkdir -p ~/.config/matplotlib
- echo "backend:template" > ~/.config/matplotlib/matplotlibrc
- sed -i 's/--doctest-modules //g' pytest.ini
- env
- pip3 list
- python3 -m pytest -v -n $NUM_CORES --cov-report html --cov-report xml --cov=. --junitxml=report.xml tests/test_*vec*.py tests/test_random.py tests/test_half_precision.py
- python3 -m coverage xml
tags:
- docker
- AVX
artifacts:
when: always
paths:
- coverage_report
reports:
coverage_report:
coverage_format: cobertura
path: coverage.xml
junit: report.xml
arm64v8:
extends: .multiarch_template
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/arm64
variables:
QEMU_CPU: "cortex-a76"
before_script:
- *multiarch_before_script
- sed -i s/march=native/march=armv8-a/g ~/.config/pystencils/config.json
ppc64le:
extends: .multiarch_template
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/ppc64le
before_script:
- *multiarch_before_script
- sed -i s/mcpu=native/mcpu=power8/g ~/.config/pystencils/config.json
arm64v9:
# SVE support is still unreliable in GCC 11 (incorrect code for fixed-width vectors, internal compiler errors).
# For half precision Clang is necessary
extends: .multiarch_template
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/arm64
before_script:
- *multiarch_before_script
- sed -i s/march=native/march=armv9-a+sve2+sme/g ~/.config/pystencils/config.json
- sed -i s/g\+\+/clang++/g ~/.config/pystencils/config.json
riscv64:
# RISC-V vector extension are currently not supported by GCC.
extends: .multiarch_template
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/riscv64
variables:
# explicitly set SIMD as detection requires QEMU >= 8.1
PYSTENCILS_SIMD: "rvv"
QEMU_CPU: "rv64,v=true,zicboz=true"
before_script:
- *multiarch_before_script
- sed -i 's/march=native/march=rv64imfdvzicboz/g' ~/.config/pystencils/config.json
- sed -i s/g\+\+/clang++-15/g ~/.config/pystencils/config.json
minimal-conda:
stage: pretest
extends: .every-commit
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/minimal_conda
before_script:
- pip install -e .
script:
- python quicktest.py
tags:
- docker
- cuda
minimal-sympy-master:
stage: test
extends: .every-commit
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/minimal_conda
before_script:
- pip install -e .
script:
- python -m pip install --upgrade git+https://github.com/sympy/sympy.git
- python quicktest.py
allow_failure: true
tags:
- docker
- cuda
pycodegen-integration:
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/full
stage: test
when: manual
allow_failure: true
script:
- git clone https://gitlab-ci-token:${CI_JOB_TOKEN}@i10git.cs.fau.de/pycodegen/pycodegen.git
- cd pycodegen
- git submodule sync --recursive
- git submodule update --init --recursive
- git submodule foreach git fetch origin # compare the latest master version!
- git submodule foreach git reset --hard origin/master
- cd pystencils
- git remote add test $CI_REPOSITORY_URL
- git fetch test
- git reset --hard $CI_COMMIT_SHA
- cd ..
- pip install -e pystencils/
- pip install -e lbmpy/
- cmake --version
- ./install_walberla.sh
- export NUM_CORES=$(nproc --all)
- mkdir -p ~/.config/matplotlib
- echo "backend:template" > ~/.config/matplotlib/matplotlibrc
- cd pystencils
- py.test -v -n $NUM_CORES --junitxml=report.xml .
- cd ../lbmpy
- py.test -v -n $NUM_CORES --junitxml=report.xml .
- cd ../walberla/build/
- make -j $NUM_CORES CodegenJacobiCPU CodegenJacobiGPU CodegenPoissonCPU CodegenPoissonGPU MicroBenchmarkGpuLbm LbCodeGenerationExample
- make -j $NUM_CORES multiphaseCPU multiphaseGPU FluctuatingMRT FlowAroundSphereCodeGen FieldLayoutAndVectorizationTest GeneratedOutflowBC
- cd apps/benchmarks/UniformGridGPU
- make -j $NUM_CORES
- cd ../UniformGridCPU
- make -j $NUM_CORES
tags:
- docker
- cuda11
- AVX
artifacts:
when: always
reports:
junit: pycodegen/*/report.xml
# -------------------- Scheduled Tasks --------------------------------------------------------------------------
# Nightly test against the latest (pre-release) version of SymPy published on PyPI
nightly-sympy:
stage: nightly
needs: []
extends: .scheduled
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/latest_python
before_script:
- pip install -e .
- pip install --upgrade --pre sympy
script:
- env
- pip list
- export NUM_CORES=$(nproc --all)
- mkdir -p ~/.config/matplotlib
- echo "backend:template" > ~/.config/matplotlib/matplotlibrc
- mkdir public
- pytest -v -n $NUM_CORES -m "not longrun" --junitxml=report.xml
tags:
- docker
- AVX
- cuda
artifacts:
when: always
reports:
junit: report.xml
# -------------------- Linter & Documentation --------------------------------------------------------------------------
flake8-lint:
stage: pretest
extends: .every-commit
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/full
script:
- flake8 src/pystencils
tags:
- docker
build-documentation:
stage: docs
extends: .every-commit
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/documentation
needs: []
before_script:
- pip install -e .
script:
- mkdir html_doc
- sphinx-build -b html doc html_doc
- sphinx-build -W -b html doc html_doc
tags:
- docker
artifacts:
paths:
- html_doc
pages:
image: i10git.cs.fau.de:5005/pycodegen/pycodegen/full
extends: .every-commit-master
stage: deploy
needs: ["tests-and-coverage", "build-documentation"]
script:
- ls -l
- mv coverage_report html_doc
- mv html_doc public # folder has to be named "public" for gitlab to publish it
artifacts:
paths:
- public
tags:
- docker
[settings]
line_length=100
balanced_wrapping=True
multi_line_output=4
known_third_party=sympy
Contributors:
-------------
- Martin Bauer <martin.bauer@fau.de>
- Markus Holzer <markus.holzer@fau.de>
- Stephan Seitz <stephan.seitz@fau.de>
- Michael Kuron <mkuron@icp.uni-stuttgart.de>
- Jan Hönig <jan.hoenig@fau.de>
- Julian Hammer <julian.hammer@fau.de>
- Nils Kohl <nils.kohl@fau.de>
- Frederik Hennig <frederik.hennig@fau.de>
- Dominik Ernst <dominik.ernst@fau.de>
- Christian Godenschwager <christian.godenschwager@fau.de>
- Dominik Thoennes <dominik.thoennes@fau.de>
# Change Log
## Unreleased
### Removed
* LLVM backend because it was not used much and not good integrated in pystencils.
* OpenCL backend because it was not used much and not good integrated in pystencils.
# Contributing
Contributions to pystencils are always welcome, and they are greatly appreciated!
A list of open problems can be found [here]( https://i10git.cs.fau.de/pycodegen/pystencils/-/issues).
Of course, it is also always appreciated to bring own ideas and problems to the community!
Please submit all contributions to the official [GitLab repository](https://i10git.cs.fau.de/pycodegen/pystencils) in the form of a Merge Request. Please do not submit git diffs or files containing the changes.
There also exists a GitHub repository, which is only a mirror to the GitLab repository. Contributions to the GitHub repository are not considered.
`pystencils` is an open-source python package under the license of AGPL3. Thus we consider the act of contributing to the code by submitting a Merge Request as the "Sign off" or agreement to the AGPL3 license.
You can contribute in many different ways:
## Types of Contributions
### Report Bugs
Report bugs at [https://i10git.cs.fau.de/pycodegen/pystencils/-/issues](https://i10git.cs.fau.de/pycodegen/pystencils/-/issues).
For pystencils, it is often necessary to provide the python and [SymPy](https://www.sympy.org/en/index.html) versions used and hardware information like the
processor architecture and the compiler version used to compile the generated kernels.
### Fix Issues
Look through the GitLab issues. Different tags are indicating the status of the issues.
The "bug" tag indicates problems with pystencils, while the "feature" tag shows ideas that should be added in the future.
### Write Documentation
The documentation of pystencils can be found [here](https://pycodegen.pages.i10git.cs.fau.de/pystencils/). Jupyter notebooks are used to provide an
interactive start to pystencils. It is always appreciated if new document notebooks are provided
since this helps others a lot.
## Get Started!
Ready to contribute? Here is how to set up `pystencils` for local development.
1. Fork the `pystencils` repo on GitLab.
2. Clone your fork locally:
```bash
$ git clone https://i10git.cs.fau.de/your-name/pystencils
```
3. Install your local copy into a virtualenv. It is also recommended to use anaconda or miniconda to manage the python environments.
```bash
$ mkvirtualenv pystencils
$ cd pystencils/
$ pip install -e .
```
4. Create a branch for local development:
```bash
$ git checkout -b name-of-your-bugfix-or-feature
```
Now you can make your changes locally.
5. When you're done making changes, check that your changes pass flake8 and the
tests
```bash
$ flake8 pystencils
$ py.test -v -n $NUM_CORES -m "not longrun" .
```
To get all packages needed for development, a requirements list can be found [here](https://i10git.cs.fau.de/pycodegen/pycodegen/-/blob/master/conda_environment_dev.yml). This includes flake8 and pytest.
6. Commit your changes and push your branch to GitHub::
```bash
$ git add .
$ git commit -m "Your detailed description of your changes."
$ git push origin name-of-your-bugfix-or-feature
```
7. Submit a Merge Request on GitLab.
## Merge Request Guidelines
Before you submit a Merge Request, check that it meets these guidelines:
1. All functionality that is implemented through this Merge Request should be covered by unit tests. These are implemented in `pystencil_tests`
2. If the Merge Request adds functionality, the docs should be updated. Put your new functionality into a function with a docstring.
3. If you have a maintainer status for `pystencils`, you can merge Merge Requests to the master branch. However, every Merge Request needs to be reviewed by another developer. Thus it is not allowed to merge a Merge Request, which is submitted by oneself.
## Tips
To run a subset of tests:
```bash
$ py.test my_test.py
```
\ No newline at end of file
This diff is collapsed.
include AUTHORS.txt
include CONTRIBUTING.md
include CHANGELOG.md
pystencils
==========
[![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/mabau/pystencils/master?filepath=doc%2Fnotebooks)
[![Docs](https://img.shields.io/badge/read-the_docs-brightgreen.svg)](https://pycodegen.pages.i10git.cs.fau.de/pystencils)
[![pypi-package](https://badge.fury.io/py/pystencils.svg)](https://badge.fury.io/py/pystencils)
[![pipeline status](https://i10git.cs.fau.de/pycodegen/pystencils/badges/master/pipeline.svg)](https://i10git.cs.fau.de/pycodegen/pystencils/commits/master)
[![coverage report](https://i10git.cs.fau.de/pycodegen/pystencils/badges/master/coverage.svg)](http://pycodegen.pages.i10git.cs.fau.de/pystencils/coverage_report)
Run blazingly fast stencil codes on numpy arrays.
*pystencils* uses sympy to define stencil operations, that can be executed on numpy arrays.
Exploiting the stencil structure makes *pystencils* run faster than normal numpy code and even as Cython and numba,
[as demonstrated in this notebook](https://pycodegen.pages.i10git.cs.fau.de/pystencils/notebooks/demo_benchmark.html).
Here is a code snippet that computes the average of neighboring cells:
```python
import pystencils as ps
import numpy as np
f, g = ps.fields("f, g : [2D]")
stencil = ps.Assignment(g[0, 0],
(f[1, 0] + f[-1, 0] + f[0, 1] + f[0, -1]) / 4)
kernel = ps.create_kernel(stencil).compile()
f_arr = np.random.rand(1000, 1000)
g_arr = np.empty_like(f_arr)
kernel(f=f_arr, g=g_arr)
```
*pystencils* is mostly used for numerical simulations using finite difference or finite volume methods.
It comes with automatic finite difference discretization for PDEs:
```python
import pystencils as ps
import sympy as sp
c, v = ps.fields("c, v(2): [2D]")
adv_diff_pde = ps.fd.transient(c) - ps.fd.diffusion(c, sp.symbols("D")) + ps.fd.advection(c, v)
discretize = ps.fd.Discretization2ndOrder(dx=1, dt=0.01)
discretization = discretize(adv_diff_pde)
```
Installation
------------
```bash
pip install pystencils[interactive]
```
Without `[interactive]` you get a minimal version with very little dependencies.
All options:
- `gpu`: use this if an NVIDIA or AMD GPU is available and CUDA or ROCm is installed
- `alltrafos`: pulls in additional dependencies for loop simplification e.g. libisl
- `bench_db`: functionality to store benchmark result in object databases
- `interactive`: installs dependencies to work in Jupyter including image I/O, plotting etc.
- `doc`: packages to build documentation
Options can be combined e.g.
```bash
pip install pystencils[interactive, gpu, doc]
```
pystencils is also fully compatible with Windows machines. If working with visual studio and cupy makes sure to run example files first to ensure that cupy can find the compiler's executable.
Documentation
-------------
Read the docs [here](https://pycodegen.pages.i10git.cs.fau.de/pystencils) and
check out the Jupyter notebooks in `doc/notebooks`. The **Changelog** of pystencils can be found [here](https://i10git.cs.fau.de/pycodegen/pystencils/-/blob/master/CHANGELOG.md).
Authors
-------
Many thanks go to the [contributors](https://i10git.cs.fau.de/pycodegen/pystencils/-/blob/master/AUTHORS.txt) of pystencils.
### Please cite us
If you use pystencils in a publication, please cite the following articles:
Overview:
- M. Bauer et al, Code Generation for Massively Parallel Phase-Field Simulations. Association for Computing Machinery, 2019. https://doi.org/10.1145/3295500.3356186
Performance Modelling:
- D. Ernst et al, Analytical performance estimation during code generation on modern GPUs. Journal of Parallel and Distributed Computing, 2023. https://doi.org/10.1016/j.jpdc.2022.11.003
from pystencils.field import Field, extractCommonSubexpressions
from pystencils.typedsymbol import TypedSymbol
from pystencils.slicing import makeSlice
import sympy as sp
import textwrap as textwrap
from sympy.tensor import IndexedBase, Indexed
from pystencils.field import Field
from pystencils.typedsymbol import TypedSymbol
class Node(object):
"""Base class for all AST nodes"""
def __init__(self, parent=None):
self.parent = parent
def args(self):
"""Returns all arguments/children of this node"""
return []
@property
def symbolsDefined(self):
"""Set of symbols which are defined by this node. """
return set()
@property
def undefinedSymbols(self):
"""Symbols which are use but are not defined inside this node"""
raise NotImplementedError()
def atoms(self, argType):
"""
Returns a set of all children which are an instance of the given argType
"""
result = set()
for arg in self.args:
if isinstance(arg, argType):
result.add(arg)
result.update(arg.atoms(argType))
return result
def parents(self):
return None
def children(self):
return None
class KernelFunction(Node):
class Argument:
def __init__(self, name, dtype):
self.name = name
self.dtype = dtype
self.isFieldPtrArgument = False
self.isFieldShapeArgument = False
self.isFieldStrideArgument = False
self.isFieldArgument = False
self.fieldName = ""
self.coordinate = None
if name.startswith(Field.DATA_PREFIX):
self.isFieldPtrArgument = True
self.isFieldArgument = True
self.fieldName = name[len(Field.DATA_PREFIX):]
elif name.startswith(Field.SHAPE_PREFIX):
self.isFieldShapeArgument = True
self.isFieldArgument = True
self.fieldName = name[len(Field.SHAPE_PREFIX):]
elif name.startswith(Field.STRIDE_PREFIX):
self.isFieldStrideArgument = True
self.isFieldArgument = True
self.fieldName = name[len(Field.STRIDE_PREFIX):]
def __repr__(self):
return '<{0} {1}>'.format(self.dtype, self.name)
def __init__(self, body, functionName="kernel"):
super(KernelFunction, self).__init__()
self._body = body
self._parameters = None
self._functionName = functionName
self._body.parent = self
# these variables are assumed to be global, so no automatic parameter is generated for them
self.globalVariables = set()
@property
def symbolsDefined(self):
return set()
@property
def undefinedSymbols(self):
return set()
@property
def parameters(self):
self._updateParameters()
return self._parameters
@property
def body(self):
return self._body
@property
def args(self):
return [self._body]
@property
def functionName(self):
return self._functionName
def _updateParameters(self):
undefinedSymbols = self._body.undefinedSymbols - self.globalVariables
self._parameters = [KernelFunction.Argument(s.name, s.dtype) for s in undefinedSymbols]
self._parameters.sort(key=lambda l: (l.fieldName, l.isFieldPtrArgument, l.isFieldShapeArgument,
l.isFieldStrideArgument, l.name),
reverse=True)
def children(self):
yield self.body
def __repr__(self):
self._updateParameters()
return '{0} {1}({2})\n{3}'.format(type(self).__name__, self.functionName, self.parameters,
textwrap.indent(repr(self.body), '\t'))
class Block(Node):
def __init__(self, listOfNodes):
super(Node, self).__init__()
self._nodes = listOfNodes
for n in self._nodes:
n.parent = self
@property
def args(self):
return self._nodes
def insertFront(self, node):
node.parent = self
self._nodes.insert(0, node)
def insertBefore(self, newNode, insertBefore):
newNode.parent = self
idx = self._nodes.index(insertBefore)
self._nodes.insert(idx, newNode)
def append(self, node):
node.parent = self
self._nodes.append(node)
def takeChildNodes(self):
tmp = self._nodes
self._nodes = []
return tmp
def replace(self, child, replacements):
idx = self._nodes.index(child)
del self._nodes[idx]
if type(replacements) is list:
for e in replacements:
e.parent = self
self._nodes = self._nodes[:idx] + replacements + self._nodes[idx:]
else:
replacements.parent = self
self._nodes.insert(idx, replacements)
@property
def symbolsDefined(self):
result = set()
for a in self.args:
result.update(a.symbolsDefined)
return result
@property
def undefinedSymbols(self):
result = set()
definedSymbols = set()
for a in self.args:
result.update(a.undefinedSymbols)
definedSymbols.update(a.symbolsDefined)
return result - definedSymbols
def children(self):
yield self._nodes
def __repr__(self):
return ''.join('{!r}\n'.format(node) for node in self._nodes)
class PragmaBlock(Block):
def __init__(self, pragmaLine, listOfNodes):
super(PragmaBlock, self).__init__(listOfNodes)
self.pragmaLine = pragmaLine
class LoopOverCoordinate(Node):
LOOP_COUNTER_NAME_PREFIX = "ctr"
def __init__(self, body, coordinateToLoopOver, start, stop, step=1):
self._body = body
self._coordinateToLoopOver = coordinateToLoopOver
self._begin = start
self._end = stop
self._increment = step
self._body.parent = self
self.prefixLines = []
def newLoopWithDifferentBody(self, newBody):
result = LoopOverCoordinate(newBody, self._coordinateToLoopOver, self._begin, self._end, self._increment)
result.prefixLines = self.prefixLines
return result
@property
def args(self):
result = [self._body]
for e in [self._begin, self._end, self._increment]:
if hasattr(e, "args"):
result.append(e)
return result
@property
def body(self):
return self._body
@property
def start(self):
return self._begin
@property
def stop(self):
return self._end
@property
def step(self):
return self._increment
@property
def coordinateToLoopOver(self):
return self._coordinateToLoopOver
@property
def symbolsDefined(self):
return set([self.loopCounterSymbol])
@property
def undefinedSymbols(self):
result = self._body.undefinedSymbols
for possibleSymbol in [self._begin, self._end, self._increment]:
if isinstance(possibleSymbol, Node) or isinstance(possibleSymbol, sp.Basic):
result.update(possibleSymbol.atoms(sp.Symbol))
return result - set([self.loopCounterSymbol])
@staticmethod
def getLoopCounterName(coordinateToLoopOver):
return "%s_%s" % (LoopOverCoordinate.LOOP_COUNTER_NAME_PREFIX, coordinateToLoopOver)
@property
def loopCounterName(self):
return LoopOverCoordinate.getLoopCounterName(self.coordinateToLoopOver)
@staticmethod
def getLoopCounterSymbol(coordinateToLoopOver):
return TypedSymbol(LoopOverCoordinate.getLoopCounterName(coordinateToLoopOver), "int")
@property
def loopCounterSymbol(self):
return LoopOverCoordinate.getLoopCounterSymbol(self.coordinateToLoopOver)
@property
def isOutermostLoop(self):
from pystencils.transformations import getNextParentOfType
return getNextParentOfType(self, LoopOverCoordinate) is None
@property
def isInnermostLoop(self):
return len(self.atoms(LoopOverCoordinate)) == 0
@property
def coordinateToLoopOver(self):
return self._coordinateToLoopOver
def children(self):
return self.body
def __repr__(self):
return 'loop:{!s} in {!s}:{!s}:{!s}\n{!s}'.format(self.loopCounterName, self.start, self.stop, self.step,
textwrap.indent(repr(self.body), '\t'))
class SympyAssignment(Node):
def __init__(self, lhsSymbol, rhsTerm, isConst=True):
self._lhsSymbol = lhsSymbol
self.rhs = rhsTerm
self._isDeclaration = True
if isinstance(self._lhsSymbol, Field.Access) or isinstance(self._lhsSymbol, IndexedBase):
self._isDeclaration = False
self._isConst = isConst
@property
def lhs(self):
return self._lhsSymbol
@lhs.setter
def lhs(self, newValue):
self._lhsSymbol = newValue
self._isDeclaration = True
if isinstance(self._lhsSymbol, Field.Access) or isinstance(self._lhsSymbol, Indexed):
self._isDeclaration = False
@property
def args(self):
return [self._lhsSymbol, self.rhs]
@property
def symbolsDefined(self):
if not self._isDeclaration:
return set()
return set([self._lhsSymbol])
@property
def undefinedSymbols(self):
result = self.rhs.atoms(sp.Symbol)
result.update(self._lhsSymbol.atoms(sp.Symbol))
return result
@property
def isDeclaration(self):
return self._isDeclaration
@property
def isConst(self):
return self._isConst
def __repr__(self):
return repr(self.lhs) + " = " + repr(self.rhs)
class TemporaryMemoryAllocation(Node):
def __init__(self, typedSymbol, size):
self.symbol = typedSymbol
self.size = size
@property
def symbolsDefined(self):
return set([self.symbol])
@property
def undefinedSymbols(self):
if isinstance(self.size, sp.Basic):
return self.size.atoms(sp.Symbol)
else:
return set()
@property
def args(self):
return [self.symbol]
class TemporaryMemoryFree(Node):
def __init__(self, typedSymbol):
self.symbol = typedSymbol
@property
def symbolsDefined(self):
return set()
@property
def undefinedSymbols(self):
return set()
@property
def args(self):
return []
import textwrap
from sympy.utilities.codegen import CCodePrinter
from pystencils.ast import Node
def generateC(astNode):
"""
Prints the abstract syntax tree as C function
"""
printer = CBackend(cuda=False)
return printer(astNode)
def generateCUDA(astNode):
printer = CBackend(cuda=True)
return printer(astNode)
# --------------------------------------- Backend Specific Nodes -------------------------------------------------------
class CustomCppCode(Node):
def __init__(self, code, symbolsRead, symbolsDefined):
self._code = "\n" + code
self._symbolsRead = set(symbolsRead)
self._symbolsDefined = set(symbolsDefined)
@property
def code(self):
return self._code
@property
def args(self):
return []
@property
def symbolsDefined(self):
return self._symbolsDefined
@property
def undefinedSymbols(self):
return self.symbolsDefined - self._symbolsRead
class PrintNode(CustomCppCode):
def __init__(self, symbolToPrint):
code = '\nstd::cout << "%s = " << %s << std::endl; \n' % (symbolToPrint.name, symbolToPrint.name)
super(PrintNode, self).__init__(code, symbolsRead=[symbolToPrint], symbolsDefined=set())
# ------------------------------------------- Printer ------------------------------------------------------------------
class CBackend:
def __init__(self, cuda=False):
self.cuda = cuda
self.sympyPrinter = CustomSympyPrinter()
self._indent = " "
def __call__(self, node):
return str(self._print(node))
def _print(self, node):
for cls in type(node).__mro__:
methodName = "_print_" + cls.__name__
if hasattr(self, methodName):
return getattr(self, methodName)(node)
raise NotImplementedError("CBackend does not support node of type " + cls.__name__)
def _print_KernelFunction(self, node):
functionArguments = ["%s %s" % (s.dtype, s.name) for s in node.parameters]
prefix = "__global__ void" if self.cuda else "void"
funcDeclaration = "%s %s(%s)" % (prefix, node.functionName, ", ".join(functionArguments))
body = self._print(node.body)
return funcDeclaration + "\n" + body
def _print_Block(self, node):
blockContents = "\n".join([self._print(child) for child in node.args])
return "{\n%s\n}" % (textwrap.indent(blockContents, self._indent))
def _print_PragmaBlock(self, node):
return "%s\n%s" % (node.pragmaLine, self._print_Block(node))
def _print_LoopOverCoordinate(self, node):
counterVar = node.loopCounterName
start = "int %s = %s" % (counterVar, self.sympyPrinter.doprint(node.start))
condition = "%s < %s" % (counterVar, self.sympyPrinter.doprint(node.stop))
update = "%s += %s" % (counterVar, self.sympyPrinter.doprint(node.step),)
loopStr = "for (%s; %s; %s)" % (start, condition, update)
prefix = "\n".join(node.prefixLines)
if prefix:
prefix += "\n"
return "%s%s\n%s" % (prefix, loopStr, self._print(node.body))
def _print_SympyAssignment(self, node):
dtype = ""
if node.isDeclaration:
if node.isConst:
dtype = "const " + node.lhs.dtype + " "
else:
dtype = node.lhs.dtype + " "
return "%s %s = %s;" % (dtype, self.sympyPrinter.doprint(node.lhs), self.sympyPrinter.doprint(node.rhs))
def _print_TemporaryMemoryAllocation(self, node):
return "%s * %s = new %s[%s];" % (node.symbol.dtype, self.sympyPrinter.doprint(node.symbol),
node.symbol.dtype, self.sympyPrinter.doprint(node.size))
def _print_TemporaryMemoryFree(self, node):
return "delete [] %s;" % (self.sympyPrinter.doprint(node.symbol),)
def _print_CustomCppCode(self, node):
return node.code
# ------------------------------------------ Helper function & classes -------------------------------------------------
class CustomSympyPrinter(CCodePrinter):
def _print_Pow(self, expr):
"""Don't use std::pow function, for small integer exponents, write as multiplication"""
if expr.exp.is_integer and expr.exp.is_number and 0 < expr.exp < 8:
return '(' + '*'.join(["(" + self._print(expr.base) + ")"] * expr.exp) + ')'
else:
return super(CustomSympyPrinter, self)._print_Pow(expr)
def _print_Rational(self, expr):
"""Evaluate all rationals i.e. print 0.25 instead of 1.0/4.0"""
return str(expr.evalf().num)
def _print_Equality(self, expr):
"""Equality operator is not printable in default printer"""
return '((' + self._print(expr.lhs) + ") == (" + self._print(expr.rhs) + '))'
def _print_Piecewise(self, expr):
"""Print piecewise in one line (remove newlines)"""
result = super(CustomSympyPrinter, self)._print_Piecewise(expr)
return result.replace("\n", "")
\ No newline at end of file
import llvmlite.ir as ir
import llvmlite.binding as llvm
import logging.config
from sympy.utilities.codegen import CCodePrinter
from pystencils.ast import Node
from sympy.printing.printer import Printer
from sympy import S
# S is numbers?
def generateLLVM(astNode):
return None
class LLVMPrinter(Printer):
"""Convert expressions to LLVM IR"""
def __init__(self, module, builder, fn, *args, **kwargs):
self.func_arg_map = kwargs.pop("func_arg_map", {})
super(LLVMPrinter, self).__init__(*args, **kwargs)
self.fp_type = ir.DoubleType()
#self.integer = ir.IntType(64)
self.module = module
self.builder = builder
self.fn = fn
self.ext_fn = {} # keep track of wrappers to external functions
self.tmp_var = {}
def _add_tmp_var(self, name, value):
self.tmp_var[name] = value
def _print_Number(self, n, **kwargs):
return ir.Constant(self.fp_type, float(n))
def _print_Integer(self, expr):
return ir.Constant(self.fp_type, float(expr.p))
def _print_Symbol(self, s):
val = self.tmp_var.get(s)
if not val:
# look up parameter with name s
val = self.func_arg_map.get(s)
if not val:
raise LookupError("Symbol not found: %s" % s)
return val
def _print_Pow(self, expr):
base0 = self._print(expr.base)
if expr.exp == S.NegativeOne:
return self.builder.fdiv(ir.Constant(self.fp_type, 1.0), base0)
if expr.exp == S.Half:
fn = self.ext_fn.get("sqrt")
if not fn:
fn_type = ir.FunctionType(self.fp_type, [self.fp_type])
fn = ir.Function(self.module, fn_type, "sqrt")
self.ext_fn["sqrt"] = fn
return self.builder.call(fn, [base0], "sqrt")
if expr.exp == 2:
return self.builder.fmul(base0, base0)
exp0 = self._print(expr.exp)
fn = self.ext_fn.get("pow")
if not fn:
fn_type = ir.FunctionType(self.fp_type, [self.fp_type, self.fp_type])
fn = ir.Function(self.module, fn_type, "pow")
self.ext_fn["pow"] = fn
return self.builder.call(fn, [base0, exp0], "pow")
def _print_Mul(self, expr):
nodes = [self._print(a) for a in expr.args]
e = nodes[0]
for node in nodes[1:]:
e = self.builder.fmul(e, node)
return e
def _print_Add(self, expr):
nodes = [self._print(a) for a in expr.args]
e = nodes[0]
for node in nodes[1:]:
e = self.builder.fadd(e, node)
return e
# TODO - assumes all called functions take one double precision argument.
# Should have a list of math library functions to validate this.
def _print_Function(self, expr):
name = expr.func.__name__
e0 = self._print(expr.args[0])
fn = self.ext_fn.get(name)
if not fn:
fn_type = ir.FunctionType(self.fp_type, [self.fp_type])
fn = ir.Function(self.module, fn_type, name)
self.ext_fn[name] = fn
return self.builder.call(fn, [e0], name)
def emptyPrinter(self, expr):
raise TypeError("Unsupported type for LLVM JIT conversion: %s"
% type(expr))
class Eval(object):
def __init__(self):
llvm.initialize()
llvm.initialize_all_targets()
llvm.initialize_native_target()
llvm.initialize_native_asmprinter()
self.target = llvm.Target.from_default_triple()
def compile(self, module):
logger.debug('=============Preparse')
logger.debug(str(module))
llvmmod = llvm.parse_assembly(str(module))
llvmmod.verify()
logger.debug('=============Function in IR')
logger.debug(str(llvmmod))
# TODO cpu, features, opt
cpu = llvm.get_host_cpu_name()
features = llvm.get_host_cpu_features()
logger.debug('=======Things')
logger.debug(cpu)
logger.debug(features.flatten())
target_machine = self.target.create_target_machine(cpu=cpu, features=features.flatten(), opt=2)
logger.debug('Machine = ' + str(target_machine.target_data))
with open('gen.ll', 'w') as f:
f.write(str(llvmmod))
optimize = True
if optimize:
pmb = llvm.create_pass_manager_builder()
pmb.opt_level = 2
pmb.disable_unit_at_a_time = False
pmb.loop_vectorize = True
pmb.slp_vectorize = True
# TODO possible to pass for functions
pm = llvm.create_module_pass_manager()
pm.add_instruction_combining_pass()
pm.add_function_attrs_pass()
pm.add_constant_merge_pass()
pm.add_licm_pass()
pmb.populate(pm)
pm.run(llvmmod)
logger.debug("==========Opt")
logger.debug(str(llvmmod))
with open('gen_opt.ll', 'w') as f:
f.write(str(llvmmod))
with llvm.create_mcjit_compiler(llvmmod, target_machine) as ee:
ee.finalize_object()
logger.debug('==========Machine code')
logger.debug(target_machine.emit_assembly(llvmmod))
with open('gen.S', 'w') as f:
f.write(target_machine.emit_assembly(llvmmod))
with open('gen.o', 'wb') as f:
f.write(target_machine.emit_object(llvmmod))
# fptr = CFUNCTYPE(c_double, c_double, c_double)(ee.get_function_address('add2'))
# result = fptr(2, 3)
# print(result)
return 0
if __name__ == "__main__":
logger = logging.getLogger(__name__)
else:
logger = logging.getLogger(__name__)
{
"version" : 1,
"disable_existing_loggers" : false,
"formatters" : {
"simple" :{
"format" : "[%(levelname)s]: %(message)s"
}
},
"handlers" : {
"console": {
"class": "logging.StreamHandler",
"level": "INFO",
"formatter": "simple",
"stream": "ext://sys.stdout"
},
"log_file": {
"class": "logging.FileHandler",
"level": "DEBUG",
"formatter": "simple",
"filename": "gen.log",
"mode" : "w",
"encoding": "utf8"
}
},
"loggers" : {
"generator" : {
"level" : "DEBUG",
"handlers" : ["console", "log_file"]
}
}
}
\ No newline at end of file
build-essential
graphviz
ffmpeg
# ----------------------------------------------------------------------------------------------------------------------
# Environment with all dependencies to use pystencils
#
#
# Download conda at https://conda.io/miniconda.html and create this environment by running:
# conda env create -f conda_environment_user.yml
# . activate pystencils
#
# If you have CUDA or ROCm installed and want to use your GPU, uncomment the last line to install cupy
#
# ----------------------------------------------------------------------------------------------------------------------
name: pystencils
dependencies:
# Basic dependencies:
- python >= 3.8
- numpy
- sympy >= 1.1
- appdirs # to find default cache directory on each platform
- joblib # caching on hard-disk, this is optional, but if not installed lbmpy is really slow
- cython # speed up boundary list computation (optional)
- matplotlib
- imageio
- pandas
- scipy
- pip
- pip:
- islpy # used to optimize staggered kernels
- py-cpuinfo # get cpu info like cache sizes, supported vector instruction sets, ...
- graphviz # can show abstract syntax trees as formatted graphs
- ipy_table # HTML tables for jupyter notebooks
- pyevtk # VTK output for serial simulations
- blitzdb # file-based No-SQL database to store simulation results
#- cupy # add this if you have CUDA or ROCm installed
#!/bin/bash
python3 setup.py develop
exec "$@"
\ No newline at end of file
import os
import runpy
import sys
import tempfile
import warnings
import nbformat
import pytest
from nbconvert import PythonExporter
from pystencils.boundaries.createindexlist import * # NOQA
# Trigger config file reading / creation once - to avoid race conditions when multiple instances are creating it
# at the same time
from pystencils.cpu import cpujit
# trigger cython imports - there seems to be a problem when multiple processes try to compile the same cython file
# at the same time
try:
import pyximport
pyximport.install(language_level=3)
except ImportError:
pass
SCRIPT_FOLDER = os.path.dirname(os.path.realpath(__file__))
sys.path.insert(0, os.path.abspath('pystencils'))
# the Ubuntu pipeline uses an older version of pytest which uses deprecated functionality.
# This leads to many warinings in the test and coverage pipeline.
pytest_numeric_version = [int(x, 10) for x in pytest.__version__.split('.')]
pytest_numeric_version.reverse()
pytest_version = sum(x * (100 ** i) for i, x in enumerate(pytest_numeric_version))
def add_path_to_ignore(path):
if not os.path.exists(path):
return
global collect_ignore
collect_ignore += [os.path.join(SCRIPT_FOLDER, path, f) for f in os.listdir(os.path.join(SCRIPT_FOLDER, path))]
collect_ignore = [os.path.join(SCRIPT_FOLDER, "doc", "conf.py"),
os.path.join(SCRIPT_FOLDER, "src", "pystencils", "opencl", "opencl.autoinit")]
add_path_to_ignore('tests/benchmark')
add_path_to_ignore('_local_tmp')
try:
import cupy
except ImportError:
collect_ignore += [os.path.join(SCRIPT_FOLDER, "tests/test_gpu.py")]
add_path_to_ignore('src/pystencils/gpu')
try:
import waLBerla
except ImportError:
collect_ignore += [os.path.join(SCRIPT_FOLDER, "tests/test_aligned_array.py"),
os.path.join(SCRIPT_FOLDER, "tests/test_datahandling_parallel.py"),
os.path.join(SCRIPT_FOLDER, "doc/notebooks/03_tutorial_datahandling.ipynb"),
os.path.join(SCRIPT_FOLDER, "src/pystencils/datahandling/parallel_datahandling.py"),
os.path.join(SCRIPT_FOLDER, "tests/test_small_block_benchmark.ipynb")]
try:
import blitzdb
except ImportError:
add_path_to_ignore('src/pystencils/runhelper')
collect_ignore += [os.path.join(SCRIPT_FOLDER, "tests/test_parameterstudy.py")]
collect_ignore += [os.path.join(SCRIPT_FOLDER, "tests/test_json_serializer.py")]
try:
import islpy
except ImportError:
collect_ignore += [os.path.join(SCRIPT_FOLDER, "src/pystencils/integer_set_analysis.py")]
try:
import graphviz
except ImportError:
collect_ignore += [os.path.join(SCRIPT_FOLDER, "src/pystencils/backends/dot.py")]
collect_ignore += [os.path.join(SCRIPT_FOLDER, "doc/notebooks/01_tutorial_getting_started.ipynb")]
try:
import pyevtk
except ImportError:
collect_ignore += [os.path.join(SCRIPT_FOLDER, "src/pystencils/datahandling/vtk.py")]
collect_ignore += [os.path.join(SCRIPT_FOLDER, 'setup.py')]
for root, sub_dirs, files in os.walk('.'):
for f in files:
if f.endswith(".ipynb") and not any(f.startswith(k) for k in ['demo', 'tutorial', 'test', 'doc']):
collect_ignore.append(f)
class IPythonMockup:
def run_line_magic(self, *args, **kwargs):
pass
def run_cell_magic(self, *args, **kwargs):
pass
def magic(self, *args, **kwargs):
pass
def __bool__(self):
return False
class IPyNbTest(pytest.Item):
def __init__(self, name, parent, code):
super(IPyNbTest, self).__init__(name, parent)
self.code = code
self.add_marker('notebook')
@pytest.mark.filterwarnings("ignore:IPython.core.inputsplitter is deprecated")
def runtest(self):
global_dict = {'get_ipython': lambda: IPythonMockup(),
'is_test_run': True}
# disable matplotlib output
exec("import matplotlib.pyplot as p; "
"p.switch_backend('Template')", global_dict)
# in notebooks there is an implicit plt.show() - if this is not called a warning is shown when the next
# plot is created. This warning is suppressed here
exec("import warnings;"
"warnings.filterwarnings('ignore', 'Adding an axes using the same arguments as a previous.*')",
global_dict)
with tempfile.NamedTemporaryFile() as f:
f.write(self.code.encode())
f.flush()
runpy.run_path(f.name, init_globals=global_dict, run_name=self.name)
class IPyNbFile(pytest.File):
def collect(self):
exporter = PythonExporter()
exporter.exclude_markdown = True
exporter.exclude_input_prompt = True
notebook_contents = self.fspath.open(encoding='utf-8')
with warnings.catch_warnings():
warnings.filterwarnings("ignore", "IPython.core.inputsplitter is deprecated")
notebook = nbformat.read(notebook_contents, 4)
code, _ = exporter.from_notebook_node(notebook)
if pytest_version >= 50403:
yield IPyNbTest.from_parent(name=self.name, parent=self, code=code)
else:
yield IPyNbTest(self.name, self, code)
def teardown(self):
pass
def pytest_collect_file(path, parent):
glob_exprs = ["*demo*.ipynb", "*tutorial*.ipynb", "test_*.ipynb"]
if any(path.fnmatch(g) for g in glob_exprs):
if pytest_version >= 50403:
return IPyNbFile.from_parent(fspath=path, parent=parent)
else:
return IPyNbFile(path, parent)